A share data

A-share Data Skill

Use this skill when the user asks for A 股数据, Chinese stock quotes, K-line/history, 东方财富 F10 financials, public announcements, news evidence, valuation inputs, or a data pack for later investment analysis.

This skill is data acquisition only. It does not issue investment advice by itself. It prepares timestamped, provenance-rich data that other research or portfolio workflows can analyze.

Core Policy

1. Real data only. Never fabricate prices, financials, valuation ratios, charts, holdings, or news. If an endpoint fails or a field is absent, record it as unavailable.
2. Record provenance. Keep URL/vendor, fetch timestamp, stock code, market, adjustment mode, date range, and any errors for every source.
3. Treat vendor fields cautiously. Public quote fields can be undocumented or change over time. Preserve raw fields when useful and label derived calculations clearly.
4. Do not blindly quote PE fields as static PE. For A-shares, vendor PE fields may be TTM, dynamic, negative because of loss-making bases, or otherwise vendor-specific. If prior-year attributable net profit and total market cap are both available, compute a rough static PE yourself as market_cap / prior_year_parent_net_profit and label it as an estimate.
5. Distinguish intraday from close. Mark intraday quotes with the vendor timestamp. Do not present an intraday quote as a closing price.
6. Respect data gaps. A missing or blocked source is a fact to report, not a reason to invent a substitute.

One-command workflow

After installing the skill with yk install a-share-data, run from the installed skill directory:

python3 scripts/a_share_data.py 301236 --market SZ --output /absolute/path/301236-data.json

Common variants:

# Infer market from code prefix and fetch the default data pack
python3 scripts/a_share_data.py 600519

# Fetch only quote and daily K-line data for a date range
python3 scripts/a_share_data.py 300750 --items quote,kline --start 2025-01-01 --end 2026-06-30

# Fetch financial/announcement evidence only
python3 scripts/a_share_data.py 301236 --items financials,cashflow,announcements --limit 10

Default output is JSON printed to stdout. Use --output to save the data pack for downstream reports.

Data sources covered

Data need	Primary public source	What to capture	Notes
Real-time quote	Sina hq.sinajs.cn	name, open, previous close, current price, high, low, volume, amount, date, time	Requires browser-like headers; usually GB18030.
Quote / valuation fields	Tencent qt.gtimg.cn	raw quote fields plus selected price/change/turnover/market-cap/PE/PB-style fields when present	Preserve raw fields; field positions are vendor-specific.
Daily K-line	Tencent web.ifzq.gtimg.cn/appstock/app/fqkline/get	date, open, close, high, low, volume, amount when available	Use qfq by default unless a different adjustment is requested.
Main financial indicators	Eastmoney F10 datacenter	revenue, parent net profit,扣非净利, margins, ROE, leverage, BPS when present	Query by SECUCODE="000000.SZ/SH/BJ"; keep raw records.
Cash flow	Eastmoney F10 datacenter	operating cash flow and related fields when present	Useful for checking profit/cash-flow divergence.
Announcements	Eastmoney announcement API	recent official announcement titles, dates, URLs/IDs when present	Prefer official announcement evidence over rumor/news.
News/search	Eastmoney search or other public search endpoints	headline, time, source, URL when body is unavailable	Treat headlines as weak signals unless the body is fetched and verified.

Code and market normalization

Accept any of these forms and normalize internally:

• 301236, 300750, 600519, 430047
• sz301236, sh600519, bj430047
• 301236.SZ, 600519.SH, 430047.BJ

Market inference default:

• 6 / 9 prefix → Shanghai (SH)
• 0 / 2 / 3 prefix → Shenzhen (SZ)
• 4 / 8 prefix → Beijing (BJ)

If inference is ambiguous or wrong for a specific instrument, pass --market SH|SZ|BJ explicitly.

Recommended data-pack shape

When saving data for a report, keep this shape:

{
  "subject": { "code": "301236", "market": "SZ", "secucode": "301236.SZ" },
  "fetched_at": "2026-06-30T...Z",
  "sources": {
    "sina_quote": { "url": "...", "data": {}, "error": null },
    "tencent_quote": { "url": "...", "data": {}, "error": null },
    "tencent_kline": { "url": "...", "data": [], "error": null },
    "eastmoney_financials": { "url": "...", "data": [], "error": null },
    "eastmoney_cashflow": { "url": "...", "data": [], "error": null },
    "eastmoney_announcements": { "url": "...", "data": [], "error": null }
  },
  "derived": [],
  "limitations": []
}

Derivation rules

Only compute derived fields when inputs are present and units are clear enough:

• Rough static PE: total_market_cap / prior_year_parent_net_profit.
• Return / drawdown / moving averages: compute from K-line closes and state adjustment mode (qfq, hfq, or raw/bfq).
• Turnover/volume interpretation: preserve vendor units; do not silently convert if units are unclear.

Every derived field must include:

• formula
• input fields and source names
• timestamp/as-of date
• limitations

Verification Checklist

After fetching data:

1. Confirm the command exits with status 0.
2. Confirm each requested source has either non-empty data or an explicit error.
3. Confirm quotes include a vendor date/time or are labeled as missing/stale.
4. Confirm K-line data range and adjustment mode match the request.
5. Confirm financial records include report dates; do not mix annual and quarterly values without labels.
6. Confirm all unavailable sources are listed in limitations or source-level error fields.
7. Before using valuation ratios in analysis, check whether the ratio is vendor-supplied or self-computed and label it accordingly.

Common Pitfalls

• Sina/Tencent endpoints may require Referer and User-Agent; retry with browser-like headers before declaring unavailable.
• Chinese endpoints often use GBK/GB18030 encoding.
• JSONP endpoints need wrapper stripping before JSON parsing.
• Eastmoney field names differ by report type and can change. Preserve raw records and map only fields you verify.
• Recent news headline search is not the same as verified announcement evidence.
• ETF/fund data uses different endpoint families; do not assume single-stock fields apply unchanged.